Somewhere in a NABTU training center last fall, a plumbing instructor sat through his first AI literacy session, a carefully designed Microsoft curriculum covering data security basics, prompt engineering fundamentals, and how to use language models to generate the lesson plans and quizzes that eat half his prep time every week, all of it delivered in the measured cadence of institutional professional development, the kind of thing that gets announced in a press release and photographed for a blog post with a dozen people smiling at laptops.
Meanwhile, in the parking lot, his 22-year-old apprentice was using ChatGPT to decode a section of the International Plumbing Code he'd been struggling with for a week.
Nobody trained him. Nobody credentialed him. He downloaded the app because the codebook is 800 pages long, the rough-in inspection was Thursday, and Google wasn't cutting it.
The Numbers Don't Add Up
In April, Microsoft and North America's Building Trades Unions expanded a partnership to bring AI literacy to the construction trades, an effort that has so far put 1,500 instructors across 50 states and Canada through a phased program covering data security, basic AI concepts, and practical applications for both classroom teaching and jobsite decision-making, with free courses now available on LinkedIn Learning that offer an industry-recognized AI credential upon completion and an extension through TradesFutures, NABTU's nonprofit arm operating in 34 states, into apprenticeship readiness programs at high schools, workforce development centers, and correctional facilities.
In June, Meta announced America's Workforce Academy, a $115 million initiative offering four weeks of classroom and hands-on instruction with guaranteed job offers for graduates, launching in Indiana, Louisiana, Ohio, and Texas.
DeWalt committed $75,000 to ABC's Trimmer Construction Education Fund for AI training grants and launched a pilot with ABC Central Florida's Innovation and Technology Center.
Add it up: more than $190 million pledged, tens of thousands of workers in the pipeline.
Do the division. 8.3 million construction workers in the United States, per the Bureau of Labor Statistics. Tens of thousands in the pipeline for formal AI education. Somewhere between 0.1% and 0.5%, depending on how generous you are with "tens of thousands." At Microsoft's current pace of 1,500 instructors per year, training every instructor in NABTU's network would take decades. That's just the instructors. The millions of workers they're supposed to teach come after.
Where Workers Actually Learn
DeWalt's "AI in the Trades" study surveyed construction professionals globally and found that 90% believe AI will be indispensable within five years, but only 8% use it today, with just 37% piloting or researching, and the rest watching from the sidelines.
Except that's not right either, because the workers who are learning about AI are learning from YouTube (40%), Coursera (39%), and video tutorials generally (42%), scrolling through "how to use ChatGPT for construction estimating" on cracked-screen phones between pours and pickups, assembling a self-taught curriculum from comment sections and ten-minute walkthroughs that nobody vetted and nobody will stand behind if the output is wrong.
NABTU's own focus groups confirmed this: apprentices told researchers they were already using AI informally to interpret electrical codes, track regulatory changes, and prep for inspections. Tom Kriger, NABTU's director of research and education, told Construction Dive: "AI can't turn wrenches and it's not going to lay bricks. AI is more likely to support decision-making and knowledge access than replace craft labor."
He's right about the wrenches. But the knowledge-access part is exactly what the apprentices figured out on their own, months before anyone handed them a syllabus.
Suffolk Put an AI Engineer in the Trailer
On June 9, Suffolk Construction announced "Jobsite of the Future," a program that embeds AI engineers directly at construction sites to sit in on project meetings and iterate on solutions for design review, procurement tracking, and payment processing, backed by more than $100 million in prior data and technology investment and a structured data lake holding 293 terabytes of construction information, the equivalent of roughly 75 billion pages of PDFs.
Suffolk is a $5 billion national contractor, and you are probably not. If you're a residential GC running $2 million custom homes in suburban Denver, you don't have a data lake. You have a filing cabinet, a superintendent who keeps the schedule in his head, and maybe a shared Dropbox that hasn't been organized since 2019. Suffolk's model proves what's possible with limitless resources. It says nothing about what's possible without them.
Where the Money Actually Goes
Meta's $115 million targets data center construction. Microsoft's NABTU partnership focuses on trades that build AI infrastructure. Strategic? Absolutely. But residential construction, where the majority of trades workers spend their careers, where the labor shortage bites hardest, where a homeowner is writing a $500,000 check and hoping the electrician got the panel right, gets almost none of it.
ABC's 2026 workforce analysis: 349,000 additional workers needed this year, with one-fifth of all electricians over 55 and a net outflow of two workers leaving for every 100 who enter. Randstad says hiring an HVAC technician now takes 56 days, two days longer than a software developer.
Data centers account for 2.3% of U.S. construction spending, per the Census Bureau. Residential and commercial account for everything else. Most of the $190 million flows to the 2.3%.
What You Should Actually Do
If you're a trades worker or apprentice, the formal programs exist and they're free. Microsoft's LinkedIn Learning courses for construction trades are open now, and completion earns an AI literacy credential that travels with your career. ABC chapters can apply for DeWalt-funded Trimmer grants to launch AI training locally, and ABC's monthly "AI Toolbox Talks" webinar series covers its 24,000 member companies.
But don't wait for the institution. Early adopters in DeWalt's study saw results: 35% reported increased productivity, 34% cited cost savings, and 35% noted improved quality control. Nearly half were using AI for site operations and planning. Not one cited a formal certification program as their starting point.
If you're running a crew: your younger workers are already using these tools. Whether you know it or not. Whether you approve or not. Ask them what they've figured out instead of pretending it isn't happening. In DeWalt's survey, 87% of workers said AI training should be embedded in trade schools, and 59% wanted hands-on exercises tied to real construction tasks. What they need is someone who knows the trade showing them how the tool applies to their specific problems on their specific jobsite, not a curriculum designed around data center wiring by a company that has never pulled a permit.
Strongest Counterargument
Institutional training programs matter even if they start small, because YouTube doesn't cover data security, doesn't cover bias in AI outputs, and doesn't cover the liability implications of submitting a permit application based on ChatGPT's interpretation of the IPC when the inspector fails the rough-in and your E&O insurance wants to know where the code analysis came from. A worker who asks ChatGPT to read the codebook and gets a wrong answer has no recourse, no framework for evaluating reliability, no institutional backstop. Microsoft's phased approach exists because the risks of unstructured adoption are real and largely invisible until they produce a failure, and an AI literacy credential, even an imperfect one, gives employers a signal and workers a floor. Scale takes time, and every training infrastructure starts somewhere. Fair enough. But the workers aren't waiting.
What This Analysis Does Not Prove
DeWalt's survey methodology, sample size, and demographic breakdown have not been fully published, which limits the generalizability of its findings. Microsoft's "1,500 instructors trained" figure does not specify the depth, duration, or assessed outcomes of that training. Meta's America's Workforce Academy has not yet produced graduates, so its effectiveness is entirely projected. Our $190 million figure combines committed first-year investments across multiple programs with different structures, timelines, and target audiences, and should not be read as a single coordinated investment. Suffolk's experience reflects the resources of a $5 billion national contractor and is not representative of the broader industry. We could not independently verify whether informal AI use among apprentices translates to better jobsite outcomes or introduces compliance risks that have not yet been documented.